Benchmark Almanac

Atrex-Bench

GPT-5.5 0.11 leads since June 2026
Not in model cards

Agents implement GPU kernels for 30 operators and 440 shapes sampled from production inference workloads. The evaluator checks compilation and numerical correctness, then scores performance against hardware roofline ceilings with weights from production GPU usage.

The June repository release predates the July technical report. Hardware, backend, fallback handling, and agent optimization settings must accompany scores. The optional optimization agent is separate from the benchmark.

Made by
Alibaba
Released
June 2026
Kind
GPU kernels
Level
Expert
Scored
Checked automatically
Answers
Code
Tasks
All public
Measures
Efficiency
Language
English
In and out
Text, Code → Code
Interaction
Single turn
Environment
Code sandbox
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GPT-5.5 #21 on the capability index
Last model added
May 2026
Scores refreshed
November 20th, 2026

In the authors’ words

Existing GPU kernel generation benchmarks draw problems from synthetic or curated sources that diverge from deployed workloads. We present Atrex-Bench, a benchmark whose 30 operators and 440 shapes are sampled directly from full-cluster production inference traces of compute-limited, memory-rich GPUs. Each problem carries an importance weight derived from its share of observed GPU time, weighted by application card-hours and computed separately for the serving phases in which it runs, together with a per-problem roofline ceiling, so the aggregate score emphasizes the kernels that consume the most serving time. Evaluating six frontier coding agents on Atrex-Bench shows that even the best vanilla model reaches only ~10% of the hardware roofline on production operators; and correctness alone overstates capability, since much of the apparent pass rate comes from PyTorch fallbacks rather than kernels the model wrote. To close this gap, we co-release Atrex-Kernel-Agent (AKA), a profile-driven kernel-optimization agent that combines iterative measure-revise search, optimization dropout for escaping stalled search contexts, and a layered GPU-optimization knowledge base (298 reference-kernel files and 244 optimization-knowledge documents, plus external upstream reference projects for API/ISA lookup). In a controlled case study, the agent converts zero-FlyDSL fallbacks into real kernels that match or exceed hand-tuned production baselines.

Abstract of Are LLM-Generated GPU Kernels Production-Ready? A Trace-Driven Benchmark and Optimization Agent (arXiv:2607.14541, CC0)
6
Models scored in its paper

GPT-5.5 leads with 0.11 among models released by May 2026

Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best.

The newest model with a score here came out in May 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-5.5 XPU-A; 30 operators 0.11
  2. 2 Claude Opus 4.7 XPU-A; 30 operators 0.06
  3. 3 Qwen3.7-Max XPU-A; 30 operators 0.05
  4. 4 Kimi K2.6 XPU-A; 30 operators 0.04
  5. 5 GLM-5.1 XPU-A; 30 operators 0.01
  6. 6 DeepSeek-V4-Pro XPU-A; 30 operators 0.01

Source: The Atrex-Bench paper

#ModelScoreSettingDateSource
1 GPT-5.5 0.11 XPU-A; 30 operators Apr 2026 Official leaderboard
2 Claude Opus 4.7 0.06 XPU-A; 30 operators Apr 2026 Official leaderboard
3 Qwen3.7-Max 0.05 XPU-A; 30 operators May 2026 Official leaderboard
4 Kimi K2.6 0.04 XPU-A; 30 operators Apr 2026 Official leaderboard
5 GLM-5.1 0.01 XPU-A; 30 operators Apr 2026 Official leaderboard
6 DeepSeek-V4-Pro 0.01 XPU-A; 30 operators Apr 2026 Official leaderboard

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Benchmark paper Not tested since Jun 2026
GPT-5.5 0.11

Source: The Atrex-Bench paper

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper
0.00 0.03 0.05 0.07 0.10 2026 0.00 0.03 0.05 0.07 0.10 2026

Source: Official Atrex-Bench leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in AI research

All 164 in AI research
  • PaperBench In use Agent replicates ICML 2024 papers from scratch. LLM judge grades against author-written hierarchical rubrics. Qwen 3.8 Max 93%
  • MLS-Bench In use Agent must invent one algorithmic change in an ML research scaffold whose gain transfers across settings. Scored relative to human baselines. Claude Fable 5.1 50.3%
  • PostTrainBench 1.2 In use CLI agent gets a base model, one H100 and a time budget to post-train it for a target benchmark. Scored on normalized gain. Claude Fable 5.1 44.6%
  • MLE-bench In use Kaggle-style ML engineering competitions. Agent trains models and is scored against competition leaderboards. GPT-6.2 Astra 96.1%
  • PMPP-Hard In use Agents write GPU kernels for 69 tasks that must be correct, and on 48 of them also run within 1.25 times the reference time. The score is the area under the solved-tasks curve over a 10-hour budget, out of 690 task-hours. Kimi K3 0.71
  • KernelBench In use Replace PyTorch operators with faster custom GPU kernels. Scored on correctness and speedup. MiniMax-M3 28.8%

Data audit

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 6 months ago, in May 2026, for Qwen3.7-Max. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the benchmark’s paper (6 entries).
  • Pass Models scored 6 models we track.
  • Warning Newest score The newest came 6 months ago, in May 2026, for Qwen3.7-Max.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.5 (#21).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 6 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check